← CanIRunLocal · All hardware · All models

Can I run Qwen3-Coder 480B-A35B (MoE) locally?

Alibaba · 480B parameters (35B active) · coding.

Ollama weight files: Q4_K_M 290 GB (qwen3-coder:480b); Q8_0 510 GB (qwen3-coder:480b-a35b-q8_0).

Yes on 5 machines in this list. The smallest is 4x Mac Studio M5 Max (40c GPU) 128GB: Q5_K_M, about 342.8 GB, roughly 7 tokens per second, via llama.cpp. llama-server -hf unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q5_K_M --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node

DGX Spark 128GB: does not fit
Mac Studio M5 Max 128GB: does not fit
Mac Studio M5 Ultra 256GB: does not fit

Setup on 4x Mac Studio M5 Max (40c GPU) 128GB

llama.cpp: llama-server -hf unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q5_K_M --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node
MLX: exo # on every Mac (brew install --cask exo); load mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit with Tensor sharding + RDMA

Open Qwen3-Coder 480B-A35B (MoE) in the checker

Hardware that fits

Hardware	Runtime	Quant	Memory	Fit	Speed	Command
4x Mac Studio M5 Max (40c GPU) 128GB	llama.cpp	Q5_K_M GGUF	342.8 GB	fits in fast memory	~7 tok/s	`llama-server -hf unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q5_K_M --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node`
2x Mac Studio M5 Ultra (80c GPU) 256GB	llama.cpp	Q5_K_M GGUF	342.8 GB	fits in fast memory	~16 tok/s	`llama-server -hf unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q5_K_M --rpc node2:50052 -c 8192 # run rpc-server on each other node`
Mac Studio M3 Ultra (80c GPU) 512GB	Ollama	Q4_K_M	292.4 GB	fits in fast memory	~15 tok/s	`ollama run qwen3-coder:480b`
4x DGX Spark 128GB	llama.cpp	Q5_K_M GGUF	342.8 GB	fits in fast memory	~5 tok/s	`llama-server -hf unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q5_K_M --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node`
4x Mac Studio M5 Ultra (80c GPU) 256GB	llama.cpp	Q8_0 GGUF	512.7 GB	fits in fast memory	~10 tok/s	`llama-server -hf unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node`

Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-16.

A project by OnDevice Labs · llms.txt